Scoring Methodology_

Philosophy

AI-assisted development makes traditional volume metrics — commits, lines of code, PR counts — increasingly meaningless. A single number that blends everything together hides the difference between a prolific code shipper and a dedicated reviewer.

Chapa replaces that single number with a multi-dimensional impact breakdown: four core dimension scores (each 0-100) plus an optional fifth Craft dimension, a developer archetype, a composite score, and a confidence rating. Each contribution style can shine on its own terms.

Normalization

Most raw metrics are transformed using logarithmic normalization to reward genuine contribution while making gaming impractical:

This produces a value between 0 and 1. The logarithmic curve means early contributions add significant value, but volume beyond the cap has zero effect. Pushing 1,000 commits does not produce a score 10x higher than 100 commits.

Signal caps

Each signal has a cap — the point beyond which additional volume adds nothing. Caps are calibrated to the P50-P75 developer range, so a solid year of consistent work can reach maximum credit.

The core dimensions

Each dimension is scored 0-100 independently. A dimension returns 0 when its primary signal is completely absent.

Delivery — shipping meaningful changes

PR weight is not a simple count. Each merged PR is weighted by its size and complexity, capped at 3.0 per PR to prevent a single massive PR from dominating. A size multiplier scales the weight from 0 to 1 as total changes (files + additions + deletions) grow from 0 to 10, so trivial or empty PRs contribute zero weight while normal PRs are unaffected.

Delivery also includes a flow efficiency modifier (±5%) based on median PR lead time — how quickly your PRs go from creation to merge. Fast turnaround (≤4 hours) earns a 5% boost; slow flow (>7 days) applies a 5% penalty. This aligns with the DORA "lead time for changes" metric.

Quality — engineering discipline

Quality is measured differently depending on your profile type. Collaborative developers (those who actively review others' code) are scored on review behavior. Solo developers — those with a review-to-PR ratio below 15% — are scored on engineering discipline signals visible in their PR workflow.

Collaborative Quality

Solo Quality

Solo developers are never penalized for working alone. Instead of skipping Quality entirely, Chapa evaluates the engineering habits visible in their PR workflow. All core dimensions are always scored for every developer.

Consistency — reliable, sustained contributions

The active days signal uses a square root curve instead of a linear ratio. This makes it easier to build early momentum — 50 active days scores 37% instead of 14% — while preserving the full range for dedicated daily contributors. Heatmap evenness uses the inverted coefficient of variation across weekly totals, with outlier weeks clipped at 3× the median to prevent a single ultra-productive week from dominating the variance. Week coverage measures what fraction of weeks had any activity at all — rewarding developers who show up consistently regardless of daily output volume.

Breadth — cross-project influence

Breadth prioritizes signals you can directly control — repo diversity (40%), contribution spread (25%), and documentation (15%) — over community signals like stars and forks that depend on external recognition. Watchers have been dropped entirely as the weakest and most passive indicator.

Developer archetypes

Your archetype is derived from the shape of your dimension profile. It tells you what kind of developer you are, not how good you are.

Tie-breaking priority: Polymath > Quality Champion > Marathoner > Builder > Artificer. If no specific archetype matches (highest dimension < 60 and not Balanced), the fallback is Emerging.

Composite score and tiers

The composite score is the average of all active dimensions, rounded to an integer. For collaborative developers, all 4 (or 5 with Craft) dimensions are included. For solo developers, Quality is excluded from the composite — only Delivery, Consistency, Breadth, and optional Craft are averaged. The score then passes through recency weighting and confidence adjustment:

The recency multiplier ranges from 0.98x (all activity is old) to 1.06x (all activity is recent), with proportional activity (25% in the last 90 days) being neutral at 1.0x. This cushions the impact of old contributions dropping out of the rolling window.

At full confidence (100), there is no confidence reduction. At minimum confidence (50), the reduction is only 7.5% — deliberate and gentle.

Confidence system

Confidence (50-100) measures signal clarity, not morality. A low confidence score never accuses wrongdoing — it simply means the data patterns make it harder to assess impact precisely.

Confidence starts at 100 and can be reduced by detected patterns:

The confidence floor is 50. No combination of penalties can push confidence below 50. All messaging is non-accusatory — we describe patterns, not intent.

Review volume imbalance and low collaboration are mutually exclusive — only one can apply at a time, so the maximum of 7 simultaneous penalties cannot be exceeded.

Score smoothing

Your displayed score uses an exponential moving average (EMA) to prevent jarring day-to-day swings. Instead of recalculating from scratch each time, the displayed score blends 15% of the newly computed score with 85% of the previous day's smoothed score.

This means a sudden 10-point raw drop manifests as roughly -1.5 per day, reaching the new level over about 4 days. On your first visit (no previous score), the raw score passes through with no smoothing.

What we deliberately exclude

Some signals are intentionally left out of scoring:

• Followers — a social metric with no correlation to engineering output
• Lines of code — easily gamed; used only for confidence heuristics, never for dimension scoring
• Private repo names — we track repo count, not identities. Your private repos are never exposed